inflation in the g7 countries: persistence and structural ...€¦ · us inflation persistence had...
TRANSCRIPT
Department of Economics and Finance
Working Paper No. 2012
http://www.brunel.ac.uk/economics
Econ
omic
s and
Fin
ance
Wor
king
Pap
er S
erie
s Guglielmo Maria Caporale, Luis Alberiko Gil-Alana and Carlos Poza
Inflation in the G7 Countries: Persistence and Structural Breaks
June 2020
1
INFLATION IN THE G7 COUNTRIES:
PERSISTENCE AND STRUCTURAL BREAKS
Guglielmo Maria Caporale, Brunel University London, United Kingdom
Luis Alberiko Gil-Alana, University of Navarra, Pamplona, Spain
and Universidad Francisco de Vitoria, Madrid, Spain
Carlos Poza, Universidad Francisco de Vitoria, Madrid, Spain
June 2020
Abstract
This paper examines long-range dependence in the inflation rates of the G7 countries by estimating their (fractional) order of integration d over the sample period January 1973 - March 2020. The results indicate that the series are very persistent, the estimated value of d being equal to or higher than 1 in all cases. Possible non-linearities in the form of Chebyshev polynomials in time are ruled out. Endogenous break tests are then carried out, and the degree of integration is estimated for each of the subsamples corresponding to the detected break dates. Significant differences are found between subsamples and countries in terms of the estimated degree of integration of the series.
Keywords: Inflation rates; G7; persistence; long memory; long-range dependence
JEL Classification: C22; E31
Corresponding author: Professor Guglielmo Maria Caporale, Department of Economics and Finance, Brunel University London, UB8 3PH, UK. Tel.: +44 (0)1895 266713. Fax: +44 (0)1895 269770. Email: [email protected]
Prof. Luis A. Gil-Alana gratefully acknowledges financial support from the MINEIC-AEI-FEDER ECO2017-85503-R project from ‘Ministerio de Economía, Industria y Competitividad’ (MINEIC), ‘Agencia Estatal de Investigación' (AEI) Spain and `Fondo Europeo de Desarrollo Regional' (FEDER). He also acknowledges support from an internal Project of the Universidad Francisco de Vitoria.
2
1. Introduction
Measuring inflation persistence is of interest to both academics (to establish the
empirical relevance of different theoretical models, such as the Phillips curve or DSGE
models) and monetary authorities (to anchor expectations in order to lower persistence
and reduce the output costs of disinflation). High persistence might result, for instance,
from price and wage rigidities (Galí and Gertler, 1999), or from the lack of transparency
of monetary policy (Walsh, 2007).
There is plenty of evidence suggesting that inflation has been highly persistent in
most developed countries since WWII (Miles et al., 2017). However, an equally
important issue is whether or not its degree of persistence has changed over time,
possibly as a result of the adoption of different monetary policy frameworks such
inflation targeting. Pivetta and Reis (2007) and Stock and Watson (2007, 2010) do not
find any significant changes in the US in the post-WWII period when accounting for
uncertainty around point estimates or distinguishing between persistent and transitory
changes in inflation. Similarly, Caporale et al. (2018) conclude that inflation persistence
has been lower in the UK in the 20th century compared to earlier ones but has not
changed significantly since WWI.
The present paper aims to provide more extensive evidence on this issue by
analysing the stochastic behaviour of inflation in all G7 countries in the last five
decades and testing for possible breaks. It uses a fractional integration framework that is
much more general and flexible than the AutoRegressive-(Integrated)-Moving Average
(AR(I)MA) models most commonly used in the literature since it is not based on the
classical I(0) versus I(1) dichotomy and allows instead the order of integration d to take
fractional as well as integer values. Having ruled out non-linearities in the series of
interest, it then carries out endogenous break tests and re-estimates d over the
3
corresponding sub-samples, thereby obtaining evidence of significant changes in
persistence across countries and subsamples.
The layout of the paper is as follows. Section 2 briefly reviews the relevant
literature. Section 3 outlines the methodology and describes the data. Section 4 presents
the empirical results. Section 5 offers some concluding remarks.
2. Literature Review
There exists an extensive literature on inflation persistence, mostly based on the
estimation of AR(I)MA models. For instance, Cogley and Sargent (2002) reported that
US inflation persistence had declined after the 1980s, whilst Pivetta and Reis (2007)
concluded that it had remained stable. Stock and Watson (2007, 2010, 2016) revisited
this issue using a model that distinguishes between transitory and permanent
components of inflation. Benati (2008) examined inflation persistence in the UK (from
1750 to 2003) and in other countries and concluded that inflation persistence cannot be
considered structural in the sense of Lucas (1976). Caporale et al. (2018) applied
fractional integration methods to UK inflation data spanning more than three centuries
and also found that persistence was higher in the most recent century but not
significantly affected by monetary policy changes. Caporale and Gil-Alana (2020)
considered an even longer period going back to 1210 and concluded that monetary and
exchange rate regime changes do not appear a significant impact on the stochastic
behaviour of inflation if one takes a long-run, historical perspective. This is in contrast
to Osborn and Sensier (2009), who had argued that both seasonal patterns and
persistence in (monthly) UK inflation had changed with the introduction of inflation
targeting in 1992, but their analysis focuses on a relatively short period and is based on
a rather restrictive ARMA modelling framework.
4
More recently, Kurozumi and Van Zandweghe (2019) have proposed a novel
theory of intrinsic inflation persistence by introducing trend inflation and variable
elasticity of demand in a DSGE model with staggered price and wage setting. With non-
zero trend inflation, variable elasticity generates intrinsic persistence in inflation
through price dispersion stemming from staggered price setting. It also introduces
intrinsic persistence in wage inflation under staggered wage setting, which affects price
inflation. Their theoretical model implies a persistent, hump-shaped response of
inflation to a monetary policy shock. Further, in their framework a credible disinflation
leads to a gradual decline in inflation and a fall in output, and lower trend inflation
reduces inflation persistence.
Correa-López et al. (2019) study the inflation process in twelve Euro Area
countries over the period 1984 – 2017 and find cross-country heterogeneity, in terms of
mean, volatility and persistence. Having estimated a wide array of unobserved
components models, they isolate trend inflation rates in a framework that allows for
time-varying inflation gap persistence and stochastic volatility in both the trend and
transitory components. A sizeable share of inflation dynamics is accounted for by
movements in the trend reflecting short-term inflation expectations, economic slack, and
openness variables. Banerjee (2017) analysed monthly consumer price inflation over the
period from January 1958 to February 2016 for 41 countries. The estimation of GARCH
(1, 1) models for the individual countries does not indicate any significant differences
between developing and developed countries in the behaviour of the conditional
volatility of inflation; however, GMM panel estimation suggests that inflation is nearly
three and half times more volatile in developing countries compared to developed
countries.
5
Concerning the emerging economies, in a recent study García and Poon (2019)
apply a Beveridge-Nelson decomposition to observed inflation rates in Asia, and
estimate a trend, or permanent component, and a transitory, or (cyclical) inflation gap.
In this context, trend inflation represents the most likely inflation rate once transitory
effects have died away and can therefore be interpreted as the optimal conditional long-
term inflation forecast. The disinflationary shocks that have hit Asia since 2014 were to
some extent transitory, and they have had asymmetric effects depending on the
behaviour of trend inflation in each country. Countries with relatively high inflation
(India, Philippines, Indonesia) benefited, and some were affected very mildly (China,
Taiwan, Hong Kong SAR, Malaysia). Among countries with inflation below target, in
those with low and constant trend inflation (Australia, New Zealand) a low inflation rate
may be long-lasting but is temporary, while in those where trend inflation has declined
(South Korea, Thailand) low inflation risks to become entrenched.
D'Amato et al. (2007) and D'Amato and Garegnani (2013) estimated high
inflation persistence in Argentina. A wider study by Capistrán and Ramos-Francia
(2006) reached similar conclusion for a wide range of Latin American countries.
Finally, Isoardi and Gil-Alana (2019) found evidence of long-memory in the inflation
rate in Argentina by applying fractionally integration methods to both monthly and
annual data (especially in the case of the former).
As for the African continent, Nyoni (2018) modelled the volatility of the
monthly inflation rate in Zimbabwe over the period from July 2009 to July 2018 and
found that an AR (1) – IGARCH (1, 1) specification is the most appropriate; they
argued that this evidence of persistence should be taken into account by monetary
authorities to design appropriate policies. High inflation persistence was also found by
Tule et al. (2020) in the case of Nigeria by estimating a fractional cointegration VAR
6
model (FCVAR - see Johansen and Nielsen, 2012) in addition to carrying out univariate
fractional integration analysis; their findings are similar for headline, core and food
inflation rates.
Moroke and Luthuli (2015) focused instead on South Africa and found evidence
of volatility clustering, leptokurtosis, asymmetric effects and non-stationarity of the
inflation series. An AR(1)_IGARCH(1,1) model appears to be the most appropriate to
capture the high degree of persistence in the conditional volatility of the series,
although it is outperformed by an AR(1)_EGARCH(2,1) specification in terms of
forecasting accuracy.
3. Methodology and Data
We use fractional integration or I(d) models which generalise the classical ARMA-
ARIMA specifications. Allowing d to take fractional as well as integer values enables
us to consider a wider range of processes including those that are mean-reverting
despite exhibiting long memory; in this case the differencing parameter is positive but
smaller than 1 and shocks have transitory but long-lasting effects.
Fractional integration and long memory processes have been related to non-
linearities by many authors (Caporale and Gil-Alana, 2007; Raggi, and Bordignon,
2012; Belmor et al., 2020); for that reason, the possibility of non-linear deterministic
trends, based on Chebyshev polynomials in time, will also be examined, still in the
context of fractional integration. Finally, given the importance of taking into account
possible breaks in the series when carrying out fractional integration analysis (Diebold
and Inoue, 2001; Granger and Hyung, 2004), we also perform endogenous break tests
and re-estimate the models over the corresponding subsamples.
7
We use monthly data on the Consumer Price Index to measure annual inflation
in the G7 countries (Canada, Japan, United States, Germany, France, Italy and UK)
from January 1973 to March 2020. The number of observations is 567, and the data
have been obtained from Refinitiv Datastream, the primary sources being the following
for each country: Canada (CANSIM – Statistics Canada), Japan (Statistics Bureau,
Ministry of Internal Affairs & Communication), USA (Bureau of Labor Statistics, U.S.
Department of Labor), Germany (Thomson Reuters), France (INSEE – National
Institute for Statistics and Economic), Italy (Istat – National Institute of Statistics) and
UK (ONS – Office for National Statistics). The series are not seasonally adjusted.
4. Empirical Results
The first estimated model is the following:
(1)
where yt is the series of interest, and ut is assumed to be a white noise and an
autocorrelated process in turn; in the latter case we use the (non-parametric) spectral
model of Bloomfield (1973) to approximate the AR structures. Following standard
practice, we consider three possible specifications, namely without deterministic terms,
with an intercept only, and with an intercept as well as a linear time trend.
TABLE 1 ABOUT HERE
Table 1 displays the estimates of d along with their corresponding 95%
confidence bands for the two cases of white noise (in the upper half) and autocorrelated
errors (in the lower half). In both cases the selected specification on the basis of the
statistical significance of the regressors includes an intercept only. The estimated values
of d under the assumption of white noise errors imply that the unit root null hypothesis
cannot be rejected for Germany (d = 0.95) and Canada (1.03); for the remaining
,...,1,0,)1(;t10ty ==-++= tuxLxt ttdbb
8
countries, they are significantly higher than 1, especially in the case of the US (1.28)
and the UK (1.38). When autocorrelated disturbances are assumed, the unit root null
cannot be rejected for the same two countries as before (Germany and Canada) as well
as for the US (1.05), whilst d is significantly higher than 1 in the other cases. Therefore,
mean reversion (d < 1) is not found in any case, which implies that shocks have
permanent effects.
The possibility of non-linear trends is investigated next by estimating the
following model:
(2)
where T is the sample size, and m indicates the order of the Chebyshev polynomials,
which are defined as:
(3)
(4)
This model was proposed in Cuestas and Gil-Alana (2016) to jointly examine
non-linearities and persistence. Note that if m = 0 the model contains only an intercept;
if m =1, it contains an intercept and a linear trend, and if m > 1 non-linearities are
present, a higher m indicating a higher degree of non-linearity. We set m = 3, θ2 and θ3
being the coefficients capturing possible non-linearities.
TABLE 2 ABOUT HERE
The results are presented in Table 2. It can be seen that the estimates of d are
very similar to those reported in Table 1; mean reversion is not found in any single case,
and the estimated values of d are equal to or higher than 1 in all cases. However, given
,...,2,1)1(,)(0
==-+å==
tuxLxtPy ttd
tm
iiTit q
,1)(,0 =tP T
( ) ...,2,1;,...,2,1,/)5.0(cos2)(, ==-= iTtTtitP Ti p
9
the lack of significance of the corresponding coefficients, there is no evidence of non-
linearities in the series (at least of the form specified above).
TABLE 3 ABOUT HERE
Therefore, next we examine the possible presence of structural breaks by
performing the tests of both Bai and Perron (2003) and Gil-Alana (2008) for multiple
breaks, the latter specifically designed for the case of fractional integration. The results
are essentially the same in both cases, which is not surprising given the previous finding
that the series exhibit unit roots, therefore we report only those for the first test in Table
3, which shows the number of breaks and their dates for each country. Three breaks are
found in the case of Italy and the UK, four in all other cases. Some of them correspond
to policy changes, for instance to the introduction of inflation targeting (IT) in the UK
in 1992 as well as in Canada, where IT had been announced in 1991, and the launch of
the Single Market Programme in Europe in 1985, which might be behind the breaks
detected around that time in France and Italy. In Germany, the dominant player in the
European Monetary System according to the German Leadership Hypothesis, a break
occurred earlier, in 1983, shortly after a new coalition government including the
CDU/CSU and FDP parties was formed under the leadership of Helmut Kohl. In Japan
the 1990s were characterised by deflation after the economic bubble burst and the 1992
break might reflect that change in economic conditions. For the US, the first break
occurred around the time of the start of the Volcker monetary regime that used interest
rates to create a nominal anchor in the form of an expected low, stable trend inflation.
TABLE 4 ABOUT HERE
10
Table 4 displays the estimates of d for each corresponding subsample and each
country under the assumption of white noise errors. For Canada, the estimated values of
d range from 0.73 in the last subsample to 1.16 in the third subsample, but the unit root
null hypothesis cannot be rejected in any single case. For France, the unit root null is
rejected in favour of d > 1 in the case of the first and fourth subsample but cannot be
rejected in the case of the others. For Germany, the unit root null cannot be rejected in
the case of the first and fourth subsamples, but it is rejected in favour of d > 1 in the
second, and mean reversion (d < 1) occurs in the third and fifth subsamples. For Italy, d
> 1 is found for the first three subsamples whilst the unit root null cannot be rejected in
the case of the fourth one. For Japan, d > 1 is found for the first and the last subsamples,
whist the unit root null cannot be rejected for the others. For the UK, d is statistically
higher than 1 in all four subsamples. Finally, for the US, d is higher than 1 in the first,
second and fourth subsamples, while the unit root null cannot be rejected for the third
and fifth subsamples.
TABLES 5 AND 6 ABOUT HERE
Table 5 reports the corresponding estimates of d under the assumption of
autocorrelated disturbances. The estimated values of d are now slightly smaller. For
example, mean reversion occurs in the case of Canada during the last two subsamples,
and also for France in the second subsample, and for Japan during the fourth subsample.
For the UK, the unit root null cannot be rejected in the case of the second and third
subsamples. Finally, for the US, the unit root null is rejected in favour of d >1 in the
first subsample, the unit root hypothesis cannot be rejected for the second and fourth
subsamples, and evidence of mean reversion (d < 1) is found for the third and fifth
subsamples.
11
5. Conclusions
This paper examines long-range dependence in the inflation rates of the G7 countries by
estimating their (fractional) order of integration d over the sample period January 1973 -
March 2020. Its key contribution is to provide extensive evidence on the issue of
whether or not inflation persistence has changed over time in the countries under
investigation.
The results indicate that the series are very persistent, the estimated value of d
being equal to or higher than 1 in all cases. Possible non-linearities in the form of
Chebychev polynomials in time are ruled out. Endogenous break tests are then carried
out, and the degree of integration is estimated for each of the subsamples corresponding
to the detected break dates. Significant differences are found between subsamples and
countries in terms of the estimated degree of integration of the series, which implies that
the degree of persistence has not remained stable over time. This is an important finding
for both academics aiming to discriminate between different theoretical models of
inflation and monetary authorities responsible for the design of appropriate stabilization
policies.
Future work will explore time variation in inflation persistence further by
explicitly modelling regime shifts in the context of a Markov-switching model and also
by allowing for gradually evolving parameters through rolling and recursive estimation
methods. In addition, it will aim to shed light on the possible determinants of
persistence by estimating a FCVAR model (Johansen and Nielsen, 2012) including
appropriate macroeconomic variables in addition to inflation.
12
References
Bai, J. and Perron, P. (2003). Computation and analysis of multiple structural change models. Journal of Applied Econometrics, 18, 1-22.
Banerjee, S. (2017). Empirical Regularities of Inflation Volatility: Evidence from Advanced and Developing Countries. South Asian Journal of Macroeconomics and Public Finance, 6 (1): 133 – 156.
Belmor, S.; Ravichandran, C. and Jarad, F. (2020). Nonlinear generalized fractional differential equations with generalized fractional integral conditions. Journal of Taibat, University of Science 14, 1.
Benati, L. (2008). Investigating Inflation Persistence Across Monetary Regimes. The Quarterly Journal of Economics, 123 (3), 1005-1060.
Capistrán, C. and Ramos-Francia, M. (2006). Inflation Dynamics in Latin America. Working Paper 2006-11, Banco de Mexico, forthcoming in Contemporary Economic Policy.
Caporale, G.M. and Gil-Alana, L.A. (2007). Nonlinearities and Fractional Integration in the US Unemployment Rate, Oxford Bulletin of Economics and Statistics 69, 4, 521-544.
Caporale, G.M. and Gil-Alana, L.A. (2020). Fractional integration and the persistence of UK inflation, 1210-2016. Forthcoming, Economic Papers.
Caporale, G.M.; Gil-Alana, L.A. and Trani, T. (2018). On the persistence of UK inflation: a long-range dependence approach. WP no. 18-03, Department of Economics and Finance, Brunel University London.
Cogley, T., and Sargent, T.J. (2002). Evolving post-World War II U.S. inflation dynamics. NBER Macroeconomics Annual 2001, Volume 16, 331-388, National Bureau of Economic Research.
Correa-López, M.; Pacce, M. and Schlepper, K. (2019). Exploring trend inflation dynamics in Euro Area countries”. Documentos de Trabajo Nº 1909. Bank of Spain.
Cuestas, J.C. and Gil-Alana, L.A. (2016). A nonlinear approach with long range dependence based on Chebyshev polynomials. Studies in Nonlinear Dynamics and Econometrics, 20, 57-94.
D’Amato, L.; Garegnani, L. and Sotes Paladino, J. (2007). Inflation persistence and changes in the monetary regime: The Argentine case. Working Paper 2007/23, Banco Central de la República Argentina.
D'Amato, L. y Garegnani, L. (2013). ¿Cuán persistente es la inflación en Argentina?: regímenes inflacionarios y dinámica de precios en los últimos 50 años, en Dinámica inflacionaria, persistencia y formación de precios y salarios, 2013, pp 91-115. Centro de Estudios Monetarios Latinoamericanos, CEMLA.
13
Diebold, F.X. and Inoue, A. (2001). Long memory and regime switching. Journal of Econometrics, 105, 131-159.
Galí, J. and Gertler, M. (1999). Inflation dynamics: a structural econometric analysis. Journal of Monetary Economics, 44, 2, 195-222.
García, J.A. and Poon, A. (2019). Inflation trends in Asia: implications for central banks. Working Paper Series, No 2338 / December 2019. European Central Bank.
Gil-Alana, L.A. (2008). Fractional integration and structural breaks at unknown periods of time. Journal of Time Series Analysis 29, 1, 163-185.
Granger, C.W.J. and Hyung, N. (2004). Occasional structural breaks and long memory with an application to the S&P 500 absolute stock returns. Journal of Empirical Finance,11: 399-421.
Isoardi, M. and Gil-Alana, L.A. (2019). Inflation in Argentina: Analysis of Persistence Using Fractional Integration. Eastern Economic Journal (45): 204–223.
Kurozumi, T. and Van Zandweghe, W. (2019). A Theory of Intrinsic Inflation Persistence. Working Paper 19-16. August 2019. The Federal Reserve Bank of Cleveland, USA.
Lucas, R. Jr. (1976). Econometric policy evaluation: A critique. Carnegie-Rochester Conference Series on Public Policy, 1(1), 19-46.
Miles, D.K.; Panizza, U.; Reis, R. and Ubide, A.J. (2017). And yet it moves: Inflation and the Great Recession. Geneva Reports on the World Economy 19, International Center for Monetary and Banking Studies, CEPR Press.
Moroke, N. and Luthuli, A. (2015). An Optimal Generalized Autoregressive Conditional Heteroskedasticity Model for Forecasting the South African inflation volatility. Journal of Economics and Behavioral Studies, 7 (4): 134 – 149.
Nyoni, Th. (2018). Modeling and Forecasting Inflation in Zimbabwe: A Generalized Autoregressive Conditionally Heteroskedastic (GARCH) approach. MPRA Paper No. 88132, Jul 2018.
Osborn, D.R. and M. Sensier (2009). UK inflation: persistence, seasonality and monetary policy. Scottish Journal of Political Economy, 56, 1, 24-44.
Pivetta, F. and Reis, R. (2007). The persistence of inflation in the United States. Journal of Economic Dynamics and Control, 31(4), 1326-1358.
Raggi, D. and Bordignon, S. (2012). Long memory and nonlinearities in realized volatility: A Markov switching approach, Computational Statistics and Data Analysis 56, 11, 3730-3742.
Stock, J.H. and Watson, M.W. (2007). Why has U.S. inflation become harder to forecast? Journal of Money, Credit and Banking, 39(s1), 3-33.
14
Stock, J.H. and Watson, M.W. (2010). Modeling inflation after the crisis. NBER Working Papers No. 16488, National Bureau of Economic Research.
Stock, J.H. and Watson, M.W. (2016). Core Inflation and Trend Inflation. The Review of Economics and Statistics, 98(4), 770-784.
Tule, M.K.; Salisu, A.A. and Godday U.E. (2020). A test for inflation persistence in Nigeria using fractional integration & fractional cointegration techniques. Economic Modelling, Elsevier, vol. 87(C), pages 225-237.
Walsh, C.E. (2007). Optimal economic transparency. International Journal of Central Banking, 3, 1, 5-36.
15
Table 1: Estimates of d in a linear set-up
i) No autocorrelation (White noise)
Series No deterministic terms
An intercept An intercept and a linear time trend
CANADA 1.01 (0.96, 1.07) 1.03 (0.97, 1.09) 1.03 (0.97, 1.09)
FRANCE 1.31 (1.08, 1.19) 1.30 (1.23, 1.38) 1.30 (1.23, 1.38)
GERMANY 0.99 (0.93, 1.05) 0.95 (0.90, 1.01) 0.95 (0.90, 1.01)
ITALY 1.19 (1.12, 1.26) 1.20 (1.13, 1.27) 1.20 (1.13, 1.27)
JAPAN 1.17 (1.11, 1.24) 1.18 (1.12, 1.25) 1.18 (1.12, 1.25)
UK 1.29 (1.23, 1.37) 1.38 (1.32, 1.46) 1.38 (1.32, 1.46)
US 1.24 (1.17, 1.32) 1.28 (1.21, 1.37) 1.28 (1.21, 1.37)
i) Autocorrelation (Bloomfield)
Series No deterministic terms
An intercept An intercept and a linear time trend
CANADA 1.07 (0.99, 1.18) 1.05 (0.95, 1.17) 1.05 (0.95, 1.17)
FRANCE 1.19 (1.09, 1.29) 1.17 (1.06, 1.32) 1.17 (1.06, 1.32)
GERMANY 1.06 (0.96, 1.17) 1.04 (0.94, 1.15) 1.04 (0.94, 1.15)
ITALY 1.20 (1.08, 1.34) 1.20 (1.06, 1.36) 1.20 (1.06, 1.36)
JAPAN 1.12 (1.03, 1.24) 1.11 (1.00, 1.23) 1.11 (1.00, 1.23)
UK 1.28 (1.15, 1.44) 1.32 (1.18, 1.50) 1.32 (1.18, 1.50)
US 1.11 (1.01, 1.22) 1.05 (0.95, 1.19) 1.05 (0.95, 1.19) In bold, the selected models on the basis of the statistical significance of the regressors. In parenthesis, the 95% confidence bands for the values of d.
16
Table 2: Estimates of d in a non-linear set-up
i) No autocorrelation (White noise)
Series d θ1 θ2 θ3 θ4 CANADA 1.02
(0.96, 1.09) -0.7687 (-0.12)
2.6303 (0.69)
1.4003 (0.75)
0.4582 (0.37)
FRANCE 1.30 (1.23, 1.38)
-3.7629 (-0.20)
4.1764 (0.35)
2.4069 (0.55)
0.8402 (0.33)
GERMANY 0.95 (0.89, 1.01)
-0.7687 (-0.12)
1.1878 (0.65)
0.5764 (0.60)
0.3786 (0.58)
ITALY 1.20 (1.13, 1.27)
-7.7879 (-0.35)
7.0211 (0.50)
2.7920 (0.50)
1.2040 (0.35)
JAPAN 1.18 (1.11, 1.25)
-4.4216 (-0.21)
3.8091 (0.29)
2.4027 (0.46)
1.4235 (0.44)
UK 1.38 (1.31, 1.45)
-10.8124 (-0.20)
7.9332 (0.23)
3.2191 (0.28)
2.9293 (0.29)
US 1.28 (1.22, 1.37)
-4.8772 (-0.21)
3.4011 (0.23)
1.6426 (0.30)
0.8259 (0.25)
i) Autocorrelation (Bloomfield)
Series d θ1 θ2 θ3 θ4 CANADA 1.03
(0.93, 1.15) -0.7751 (-0.11)
2.6273 (0.65)
1.3791 (0.71)
0.5074 (0.39)
FRANCE 1.12 (1.04, 1.20)
-2.2395 (-0.31)
4.2100 (0.96)
2.1334 (1.12)
0.7414 (0.61)
GERMANY 1.04 (0.93, 1.14)
3.2323 (0.64)
1.2027 (0.39)
0.5339 (0.36)
0.3823 (0.39)
ITALY 1.18 (1.05, 1.27)
-2.1757 (-0.13)
4.3205 (0.41)
1.0743 (0.25)
3.2179 (1.23)
JAPAN 1.10 (0.99, 1.21)
-2.8043 (-0.21)
3.1401 (0.39)
2.3652 (0.66)
1.1785 (0.51)
UK 1.21 (1.13, 1.28)
3.3763 (0.22)
0.2039 (0.02)
0.0792 (0.02)
-4.0288 (1.72)
US 1.05 (0.94, 1.15)
-2.0915 (-0.31)
2.3485 (0.58)
0.9744 (0.51)
0.5702 (0.46)
In parenthesis, in the second column the 95% confidence band for the values of d, and in the other columns t-values.
17
Table 3: Bai and Perron’s (2003) test results
Series Number of breaks Break dates
CANADA 4 1983m5; 1992m1; 1999m8; 2012m3
FRANCE 4 1985m11; 1993m4; 2002m1; 2012m11
GERMANY 4 1983m3; 1995m2; 2006m1; 2013m2
ITALY 3 1986m1; 1996m10; 2013m3
JAPAN 4 1981m7; 1992m1; 1999m2; 2006m5
UK 3 1982m6; 1992m6; 2006m8
US 4 1982m8; 1991m10; 2004m10; 2012m4
18
Table 4: Estimates of d for each subsample with white noise errors
Series Sb. No deterministic terms
An intercept An intercept and a linear time trend
CANADA
1st 0.98 (0.89, 1.10) 1.03 (0.94, 1.14) 1.03 (0.94, 1.14) 2nd 0.81 (1.67, 1.00) 0.92 (0.77, 1.10) 0.93 (0.78, 1.10) 3rd 1.15 (0.98, 1.34) 1.16 (0.99, 1.36) 1.16 (0.99, 1.36) 4th 0.95 (0.81, 1.14) 0.97 (0.82, 1.18) 0.97 (0.82, 1.18) 5th 0.61 (0.43, 0.85) 0.73 (0.56, 1.02) 0.72 (0.51, 1.02)
FRANCE
1st 1.14 (1.05, 1.26) 1.49 (1.37, 1.65) 1.49 (1.37, 1.65) 2nd 0.82 (0.65, 1.07) 1.18 (0.98, 1.47) 1.17 (0.98, 1.45) 3rd 1.11 (0.93, 1.24) 0.92 (0.78, 1.15) 0.92 (0.78, 1.15) 4th 1.10 (0.97, 1.27) 1.18 (1.05, 1.35) 1.18 (1.05, 1.35) 5th 0.83 (0.70, 1.00) 1.03 (0.87, 1.26) 1.03 (0.87, 1.26)
GERMANY
1st 1.00 (1.89, 1.16) 1.00 (0.90, 1.13) 1.00 (0.90, 1.13) 2nd 0.96 (0.84, 1.12) 1.12 (1.01, 1.28) 1.12 (1.01, 1.27) 3rd 0.74 (0.65, 0.86) 0.70 (0.60, 0.83) 0.70 (0.60, 0.83) 4th 0.92 (0.81, 1.07) 0.97 (0.85, 1.12) 0.97 (0.85, 1.12) 5th 0.77 (0.63, 0.94) 0.75 (0.63, 0.93) 0.75 (0.63, 0.93)
ITALY
1st 1.18 (1.06, 1.31) 1.20 (1.09, 1.34) 1.20 (1.09, 1.34) 2nd 0.84 (0.72, 1.01) 1.18 (1.08, 1.32) 1.18 (1.08, 1.31) 3rd 0.98 (0.88, 1.10) 1.18 (1.09, 1.29) 1.18 (1.09, 1.29) 4th 0.91 (0.76, 1.11) 0.97 (0.83, 1.15) 0.97 (0.83, 1.15)
JAPAN
1st 1.22 (1.10, 1.39) 1.26 (1.13, 1.44) 1.25 (1.13, 1.43) 2nd 0.84 (0.74, 0.98) 0.86 (0.76, 1.00) 0.87 (0.77, 1.00) 3rd 0.87 (0.72, 1.06) 0.85 (0.71, 1.05) 0.85 (0.71, 1.05) 4th 0.76 (0.60, 1.03) 0.76 (0.57, 1.03) 0.76 (0.56, 1.03) 5th 1.17 (1.04, 1.34) 1.17 (1.04, 1.34) 1.17 (1.04, 1.34)
UK
1st 1.27 (1.12, 1.43) 1.41 (1.28, 1.58) 1.41 (1.28, 1.57) 2nd 1.00 (0.86, 1.18) 1.35 (1.22, 1.53) 1.34 (1.22, 1.52) 3rd 1.04 (0.93, 1.18) 1.16 (1.05, 1.29) 1.15 (1.05, 1.29) 4th 1.27 (1.15, 1.41) 1.38 (1.26, 1.53) 1.38 (1.26, 1.53)
US
1st 1.20 (1.10, 1.29) 1.27 (1.18, 1.39) 1.27 (1.18, 1.39) 2nd 0.96 (0.78, 1.24) 1.53 (1.30, 1.85) 1.52 (1.30, 1.85) 3rd 0.98 (0.87, 1.12) 1.00 (0.87, 1.17) 1.00 (0.87, 1.17) 4th 1.23 (1.03, 1.48) 1.37 (1.14, 1.70) 1.37 (1.14, 1.70) 5th 0.93 (0.78, 1.14) 1.09 (0.89, 1.41) 1.09 (0.89, 1.41)
In bold, the selected models on the basis of the statistical significance of the regressors. In parenthesis, the 95% confidence bands for the values of d.
19
Table 5: Estimates of d for each subsample with autocorrelated errors
Series Sb. No deterministic terms
An intercept An intercept and a linear time trend
CANADA
1st 1.05 (0.87, 1.30) 1.29 (1.08, 1.55) 1.28 (1.08, 1.55) 2nd 0.60 (0.37, 0.93) 1.05 (0.50, 1.55) 1.05 (0.55, 1.53) 3rd 1.03 (0.66, 1.53) 0.98 (0.52, 1.52) 0.98 (0.52, 1.52) 4th 0.59 (0.40, 0.85) 0.49 (0.24, 0.84) 0.49 (0.23, 0.84) 5th 0.42 (0.12, 0.66) 0.42 (0.22, 0.79) 0.24 (-0.08, 0.77)
FRANCE
1st 1.15 (0.99, 1.38) 1.36 (1.13, 1.66) 1.35 (1.14, 1.68) 2nd 0.45 (0.06, 0.84) 0.43 (0.17, 0.87) 0.53 (0.23, 0.90) 3rd 0.80 (0.60, 1.13) 0.74 (0.56, 1.02) 0.74 (0.56, 1.03) 4th 0.88 (0.65, 1.19) 0.94 (0.65, 1.31) 0.94 (0.65, 1.31) 5th 0.84 (0.61, 1.22) 1.01 (0.76, 1.68) 1.01 (0.79, 1.55)
GERMANY
1st 0.93 (0.75, 1.23) 1.01 (0.83, 1.27) 1.01 (0.83, 1.27) 2nd 0.78 (0.64, 1.02) 0.97 (0.82, 1.16) 0.97 (0.83, 1.15) 3rd 0.85 (0.61, 1.09) 0.83 (0.60, 0.12) 0.83 (0.62, 0.12) 4th 1.04 (0.79, 1.40) 1.26 (0.98, 1.66) 1.26 (0.97, 1.68) 5th 0.88 (0.59, 1.33) 0.80 (0.58, 0.15) 0.79 (0.58, 0.15)
ITALY
1st 1.17 (0.97, 1.48) 1.18 (0.94, 1.49) 1.18 (0.94, 1.51) 2nd 0.70 (0.52, 0.97) 1.31 (1.10, 1.56) 1.29 (1.10, 1.56) 3rd 0.93 (0.71, 1.26) 1.43 (1.17, 1.76) 1.44 (1.16, 1.70) 4th 0.88 (0.53, 1.39) 1.02 (0.74, 1.46) 1.02 (0.75, 1.45)
JAPAN
1st 1.15 (0.96, 1.41) 1.12 (0.92, 1.41) 1.12 (0.92, 1.41) 2nd 0.82 (0.66, 1.10) 0.89 (0.74, 1.13) 0.90 (0.75, 1.13) 3rd 0.75 (0.45, 1.15) 0.73 (0.43, 1.07) 0.75 (0.52, 1.07) 4th 0.47 (0.28, 0.76) 0.41 (0.26, 0.69) 0.38 (0.14, 0.67) 5th 0.89 (0.72, 1.18) 0.93 (0.72, 1.18) 0.93 (0.72, 1.18)
UK
1st 1.27 (1.04, 1.63) 1.37 (1.11, 1.74) 1.36 (1.11, 1.75) 2nd 0.84 (0.54, 1.21) 1.21 (0.99, 1.48) 1.21 (1.00, 1.49) 3rd 1.95 (0.72, 1.30) 1.05 (0.79, 1.38) 1.05 (0.80, 1.37) 4th 1.21 (0.91, 1.54) 1.29 (1.00, 1.68) 1.29 (1.00, 1.68)
US
1st 1.38 (1.17, 1.65) 1.45 (1.24, 1.70) 1.42 (1.24, 1.69) 2nd 0.52 (0.33, 0.81) 0.75 (0.45, 1.18) 0.75 (0.43, 1.19) 3rd 0.84 (0.69, 1.02) 0.62 (0.44, 0.86) 0.62 (0.41, 0.86) 4th 0.73 (0.50, 1.12) 0.68 (0.39, 1.13) 0.69 (0.39, 1.13) 5th 0.70 (0.45, 1.01) 0.65 (0.45, 0.97) 0.65 (0.43, 0.97)
In bold, the selected models on the basis of the statistical significance of the regressors. In parenthesis, the 95% confidence bands for the values of d.